Guest lecture for ACE510

1. Background

1.1 Project objective

Our project aims to quantify the effects of conservation tillage on crop yields (corn and soybeans) in the U.S. Corn Belt.

Conservation tillage refers toas any tillage and planting system that leaves at least 30 percent of the soil surface covered by residue after planting, according to the Conservation Technology Information Center (CTIC).

Figure 1. Different tillage types

We use fine-scale satellite image data for tillage from the OpTIS program, yet aggregated to county level, to do the analysis. Our data cover 644 counties in 12 Corn-Belt states from 2005 to 2018.

Figure 2. Adoption rates of conservation tillage for corn, 2005 – 2018

1.2 What is OpTIS?

OpTIS is an automated system to map tillage, residue cover, winter cover, and soil health practices across the U.S. Corn Belt based on remote sensing data. OpTIS was developed by Applied GeoSolutions and Dagan,Inc in collaboration with The Nature Conservancy and The Conservation Technology Information Center (CTIC). The data are publicly available (yet at state level): https://www.ctic.org/OpTIS.

1.3 Limitation and solution

OpTIS data are crop-specific, but it reports acres by previous year’s crop type category (corn, soybeans, small grains, and other crops). However, crop rotation is a common production practice in the U.S. with sizable acreage rotating between crops annually. For this reason, we are unable to directly link the tillage data with crop-specific yields directly. To address this issue, we process the data in the Cropland Data Layer (CDL) from the National Agricultural Statistics Service of the U.S. Department of Agriculture (USDA – NASS) to obtain county crop rotation matrices for the four crop categories.

CDL is a raster, geo-referenced, crop specific land cover data layer, and it can be accessed from this user interface: https://nassgeodata.gmu.edu/CropScape/.

The procedures for combining the OpTIS data with CDL data can be illustrated as follows:

Figure 3. A graphical illustration of combining CDL data with OpTIS data

2. Data challenges

The CDL data are very large. One raster file that covers the entire U.S. for one year is about 20 GB. Computers with low computer memories might fail to read the data. Assume that we can read the data, we can ensivion to do the following three steps to obtain crop rotation matrices for each county:

Step 1: Merge the nation (or state) scale CDL data with county polygon files (boundary). Extract data within each county boundary.
Step 2: Merge county-level CDL data in year t and in year t + 1; then obtain crop rotation types at each pixel point (e.g., corn - corn at pixel point X in year t/t+1).
Step 3: Sum up pixel points by crop rotation types for each county (e.g., 16,000 pixel points use corn - corn in the Champaign county). Use conversion factors to convert pixel counts to acres.

I expect to take at least two weeks to get one year of crop rotation data (with errors), assuming that my computer does not burn during the data processing. This means that I am risking 2 months of my research time for some uncertain products. The supercomputer could help though.

3. My solution (with examples)

3.1 Idea

Step 1: Use CropScape API to download county-level land use data. https://nassgeodata.gmu.edu/CropScape/devhelp/help.html
Step 2: Perform crop rotation analysis for each county (< 15 MB for each file)

All processes can be facilitated by parallel computing. And it only requires 2 days.

Parallel computing is a type of computation in which many calculations or the execution of processes are carried out simultaneously (Wikipedia). To use parallel computing, one shall divide a large problem into many small problems, and then one can facilitate computing by solving several small problems simultaneously.

Parallel computing is possible now, because modern computers can contain mutliple processors, and each processor can contain multiple cores. Each core can perform a data task.

3.2 Example

See codes

3.3 Final data products

Finally, we obtain pixel counts (converted to acres later) by crop rotation types, such as corn - corn, corn - soybeans, soybeans - wheat, for each county in the contiguous U.S., where the data are available. Below is a screenshot of our data products.

Figure 4. Screenshots of generated data

4. Crop rotation in the U.S.

We assemle all the county-level data and then find intensive corn - soybeans and soybeans - corn rotations in the center of U.S. Corn Belt. The following two figures show corn/soybeans rotations over time.

Figure 5. Percent acres rotated from corn to soybeans

Figure 6. Percent acres rotated from soybeans to corn